Market etiquette or good manners in a minefield - page 100
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
A small report on the work done. I investigated the profitability of binary patterns built on EUR/USD tick quotes for about a year. The profitability was investigated as a function of
Sergey:
which patterns showed the highest profitability for d = 4 and 5?
For d= 4: -1+1+1+1 and +1-1+1-1
For d= 5: -1+1+1+1-1-1 and +1-1+1+1+1-1
These patterns show the highest profitability, which is expressed in pips per transaction, where averaging is performed on the entire sample. The fact that the maximum profitability increases with increasing pattern size doesn't mean that the profitability of TS based on this presentation mechanism also increases. The fact is that with increasing profitability the frequency of pattern encounter of this type decreases due to geometrical growth of the number of all possible patterns. If we increase the dimensionality of the pattern by one digit (for example, from 2 segments to 3), the number of combinations increases by 2 times (from 4 to 8), and the profitability increases by 20% (see figure above). Obviously, a compromise will have to be made between forecast reliability and transaction frequency. It is possible that the most "convenient" patterns (in this sense) will be 3-links.
maybe it's hard to see, but in the flat figure it seems that the most profitable patterns are invariant to H, is this true? if so, how do you think this can be explained?
Let's take a closer look at the dependence of profitability on the steepest patterns as a function of the partition horizon - H:
The dependences are given for a 6-link pattern (fig. on the left) and for a 2-link pattern (on the right). There is, however, a dependence on H. The scale on the vertical axis is different.
For d= 4: -1+1-1+1+1 and +1-1+1-1
For d= 5: -1+1-1+1-1 and +1-1+1+1-1
This is, imho, the "figure" of a flat, which by general estimate takes 85% of the time.
This, imho, is a typical answer to a typical question of Mathematics raised in one of the posts: "Which is better: forty times over once or all forty times once?
or
two market models: Cherkizovsky and the boutique on Kutuzovsky - the Porter curve in management.
The dependencies are given for the 6-link pattern (fig. left) and for the 2-link pattern (right). There is, however, a dependency on H. The scale on the vertical axis is different.
Can we assume that the difference of the sub-integral areas is the trend over the period?
If "yes", then at the highest yield of alternating patterns we have a well-known strategy: to find the most flat pair and ... "further on" (or options).
What other conclusions can be drawn?
Can we assume that the difference of the sub-integral areas is a trend over the period?
It's a bit simpler than that.
You are right, the difference between the areas under the two dependencies, will give the contribution of the trend component. But a trend is not a trend! We can distinguish two groups of trends. The first "stochastic" group includes all trends that cannot be identified statistically in one way or another. Such trends, for example, include trends in the Wiener process - they are present on history, but one cannot make profit out of them. The second type includes the so-called "deterministic" trends, or trends that can be detected on the right side of BP in the process of their formation. Such trends include sequences of ascending or descending sections of BP, whose coefficient of mutual correlation between samples in the first difference is positive.
So, stochastic trends will result in different area under the curves on the given graphs:
And deterministic trends will "evenly" decrease the yield (see fig. elipse). Now, if the lines were to "switch places" at this point, we could talk about a real trnedirectional behaviour of the quotient in the given trading horizon H.
Sergey, please attach tables with rules sorted by support and interest separately for d = 4 and 5.
It is interesting to see the effect of parity pattern on %%.
That is, present the 3-dimensional pictures I posted on the previous page as tables?
That is, present the 3-dimensional pictures I posted on the previous page in the form of tables?
I see that you haven't read the rules, although you agree with me :)
It's well laid out here http://www.basegroup.ru/library/analysis/association_rules/intro/
Briefly:
It is not only the rate of decrease of support with increasing d that is of interest, but also the dynamics of change in the interestingness of the rule.
It may be possible to find nuances in the sign-variability.
I'm used to look at the table, as you can filter and sort them in different ways, but you can also look at the graph.
I have data on the frequency of patterns (fig. right) and their predictive validity (fig. left) as a function of H:
The data are given for d=5. The red colour shows the larger values, the blue one the smaller values.
to Neutron
Unfortunately the circumstances of my life are such that I have to leave the market and the forum for an indefinite period. In light of your recent findings, I have one idea:
The idea is to detect the switching pattern +H/H and skip one to two PT counts, after each committed transaction. Definitely, there should be duration statistics(lifetime) for +H and -H strategies. A strategy should be changed after n RT samples from +H to -H(after 1 step pause) and vice versa after n RT samples from -H to +H. According to my observations of kagi - tick series splits, there is one stable and repeating constantly pattern: when the top of previous RT falls into a delta neighborhood (not more than 3-5 pips) of the current (last) Kagi top - it needs to change the strategy from +H to -H, and to catch this pattern we need to skip 1-2 RT samples after the transaction - not trade on them, but analyze.
P.S.
Thanks a lot for the science! Have a nice trend and big profits.